How it works
From sample input to hardening checklist
Aligned with OWASP LLM01:2025. Not a formal pentest.
1
Describe & sample
Describe your AI app and optionally paste untrusted user input that should never rewrite system instructions.
2
Deterministic scan
Versioned ruleset prompt-injection@2026-07-19.2 flags injection/jailbreak patterns with concrete hits.
3
Harden & review
Use the checklist. Decision-support only — not a formal pentest; does not guarantee 100% detection.